summaryrefslogtreecommitdiffstats
path: root/cuda/2d/util.cu
diff options
context:
space:
mode:
authorWillem Jan Palenstijn <Willem.Jan.Palenstijn@cwi.nl>2021-11-22 14:43:07 +0100
committerWillem Jan Palenstijn <wjp@usecode.org>2021-11-26 12:09:09 +0100
commitbe2da43a560a7241c56e727fb481f1389e9f7fdf (patch)
tree318df874ff0efb63d01749006f09318289d0844b /cuda/2d/util.cu
parentc11ee1cffd53bdf4530eb4381bf484cd1a1ce9cb (diff)
downloadastra-be2da43a560a7241c56e727fb481f1389e9f7fdf.tar.gz
astra-be2da43a560a7241c56e727fb481f1389e9f7fdf.tar.bz2
astra-be2da43a560a7241c56e727fb481f1389e9f7fdf.tar.xz
astra-be2da43a560a7241c56e727fb481f1389e9f7fdf.zip
De-duplicate 2D texture object creation
Diffstat (limited to 'cuda/2d/util.cu')
-rw-r--r--cuda/2d/util.cu60
1 files changed, 60 insertions, 0 deletions
diff --git a/cuda/2d/util.cu b/cuda/2d/util.cu
index ac360f0..2ad3c0f 100644
--- a/cuda/2d/util.cu
+++ b/cuda/2d/util.cu
@@ -126,6 +126,66 @@ void duplicateProjectionData(float* D_dst, float* D_src, unsigned int pitch, con
cudaMemcpy2D(D_dst, sizeof(float)*pitch, D_src, sizeof(float)*pitch, sizeof(float)*dims.iProjDets, dims.iProjAngles, cudaMemcpyDeviceToDevice);
}
+bool createTextureObject2D(float* data, cudaArray*& dataArray, cudaTextureObject_t& texObj, unsigned int pitch, unsigned int width, unsigned int height)
+{
+ // TODO: For very small sizes (roughly <=512x128) with few angles (<=180)
+ // not using an array is more efficient.
+
+ cudaChannelFormatDesc channelDesc =
+ cudaCreateChannelDesc(32, 0, 0, 0, cudaChannelFormatKindFloat);
+
+ dataArray = 0;
+ cudaMallocArray(&dataArray, &channelDesc, width, height);
+ cudaMemcpy2DToArray(dataArray, 0, 0, data, pitch*sizeof(float), width*sizeof(float), height, cudaMemcpyDeviceToDevice);
+
+ cudaResourceDesc resDesc;
+ memset(&resDesc, 0, sizeof(resDesc));
+ resDesc.resType = cudaResourceTypeArray;
+ resDesc.res.array.array = dataArray;
+
+ cudaTextureDesc texDesc;
+ memset(&texDesc, 0, sizeof(texDesc));
+ texDesc.addressMode[0] = cudaAddressModeBorder;
+ texDesc.addressMode[1] = cudaAddressModeBorder;
+ texDesc.filterMode = cudaFilterModeLinear;
+ texDesc.readMode = cudaReadModeElementType;
+ texDesc.normalizedCoords = 0;
+
+ texObj = 0;
+
+ return checkCuda(cudaCreateTextureObject(&texObj, &resDesc, &texDesc, NULL), "createTextureObject2D");
+}
+
+bool createTextureObjectPitch2D(float* data, cudaTextureObject_t& texObj, unsigned int pitch, unsigned int width, unsigned int height, cudaTextureAddressMode mode)
+{
+ cudaChannelFormatDesc channelDesc =
+ cudaCreateChannelDesc(32, 0, 0, 0, cudaChannelFormatKindFloat);
+
+ cudaResourceDesc resDesc;
+ memset(&resDesc, 0, sizeof(resDesc));
+ resDesc.resType = cudaResourceTypePitch2D;
+ resDesc.res.pitch2D.devPtr = (void*)data;
+ resDesc.res.pitch2D.desc = channelDesc;
+ resDesc.res.pitch2D.width = width;
+ resDesc.res.pitch2D.height = height;
+ resDesc.res.pitch2D.pitchInBytes = sizeof(float)*pitch;
+
+ cudaTextureDesc texDesc;
+ memset(&texDesc, 0, sizeof(texDesc));
+ texDesc.addressMode[0] = mode;
+ texDesc.addressMode[1] = mode;
+ texDesc.filterMode = cudaFilterModeLinear;
+ texDesc.readMode = cudaReadModeElementType;
+ texDesc.normalizedCoords = 0;
+
+ texObj = 0;
+
+ return checkCuda(cudaCreateTextureObject(&texObj, &resDesc, &texDesc, NULL), "createTextureObjectPitch2D");
+}
+
+
+
+
template <unsigned int blockSize>
__global__ void reduce1D(float *g_idata, float *g_odata, unsigned int n)
{